utils/models.py

import re

import openai
import pandas as pd
import streamlit_scrollable_textbox as stx
import torch
from InstructorEmbedding import INSTRUCTOR
from gradio_client import Client
from transformers import (
    AutoModelForMaskedLM,
    AutoTokenizer,
)
from rank_bm25 import BM25Okapi, BM25L, BM25Plus
import numpy as np
from nltk.tokenize import word_tokenize
from nltk.corpus import stopwords
from nltk.stem.porter import PorterStemmer
import re
import streamlit as st


@st.cache_resource
def get_data():
    data = pd.read_csv("earnings_calls_cleaned_metadata_keywords_indices.csv")
    return data


# Preprocessing for BM25


def tokenizer(
    string, reg="[a-zA-Z'-]+|[0-9]{1,}%|[0-9]{1,}\.[0-9]{1,}%|\d+\.\d+%}"
):
    regex = reg
    string = string.replace("-", " ")
    return " ".join(re.findall(regex, string))


def preprocess_text(text):
    # Convert to lowercase
    text = text.lower()
    # Tokenize the text
    tokens = word_tokenize(text)
    # Remove stop words
    stop_words = set(stopwords.words("english"))
    tokens = [token for token in tokens if token not in stop_words]
    # Stem the tokens
    porter_stemmer = PorterStemmer()
    tokens = [porter_stemmer.stem(token) for token in tokens]
    # Join the tokens back into a single string
    preprocessed_text = " ".join(tokens)
    preprocessed_text = tokenizer(preprocessed_text)

    return preprocessed_text


# Initialize models from HuggingFace


@st.cache_resource
def get_splade_sparse_embedding_model():
    model_sparse = "naver/splade-cocondenser-ensembledistil"
    # check device
    device = "cuda" if torch.cuda.is_available() else "cpu"
    tokenizer = AutoTokenizer.from_pretrained(model_sparse)
    model_sparse = AutoModelForMaskedLM.from_pretrained(model_sparse)
    # move to gpu if available
    model_sparse.to(device)
    return model_sparse, tokenizer


@st.cache_resource
def get_instructor_embedding_model():
    model = INSTRUCTOR("hkunlp/instructor-xl")
    return model


@st.cache_resource
def get_instructor_embedding_model_api():
    client = Client("https://awinml-api-instructor-xl-2.hf.space/")
    return client


@st.cache_resource
def get_alpaca_model():
    client = Client("https://awinml-alpaca-cpp.hf.space")
    return client


@st.cache_resource
def get_vicuna_ner_1_model():
    client = Client("https://awinml-api-vicuna-openblas-ner-1.hf.space/")
    return client


@st.cache_resource
def get_vicuna_ner_2_model():
    client = Client("https://awinml-api-vicuna-openblas-ner-2.hf.space/")
    return client


@st.cache_resource
def get_vicuna_text_gen_model():
    client = Client("https://awinml-api-vicuna-openblas-4.hf.space/")
    return client


@st.cache_resource
def get_bm25_model(data):
    corpus = data.Text.tolist()
    corpus_clean = [preprocess_text(x) for x in corpus]
    tokenized_corpus = [doc.split(" ") for doc in corpus_clean]
    bm25 = BM25Plus(tokenized_corpus)
    return corpus, bm25


@st.cache_resource
def save_key(api_key):
    return api_key


# Text Generation


def vicuna_text_generate(prompt, model):
    generated_text = model.predict(prompt, api_name="/predict")
    return generated_text


def gpt_turbo_model(prompt):
    response = openai.ChatCompletion.create(
        model="gpt-3.5-turbo",
        messages=[
            {"role": "user", "content": prompt},
        ],
        temperature=0.01,
        max_tokens=1024,
    )
    return response["choices"][0]["message"]["content"]